Announcing Confidential Computing General Access on NVIDIA H100 Tensor Core GPUs
Table of Contents
- Introduction
- New Features
- The Importance of Security
- Hardware and Software Security for NVIDIA H100 GPUs
- Summary
Introduction
NVIDIA has announced the general access of Confidential Computing on H100 Tensor Core GPUs, which was initially launched in private preview in July 2023. This solution provides the highest level of security with optimal performance, particularly for virtualized environments.
New Features
- Support for single-GPU passthrough with Confidential Computing is now generally available on CUDA 12.4.
- Performance optimizations have been introduced.
- Security hardening of the driver specific to Confidential Computing.
The Importance of Security
In today's interconnected world, data security is crucial. With the rise of AI workloads, particularly concerning the protection of data and code, Confidential Computing becomes essential. This is especially true for sensitive data containing PII or IP, where running AI models in the cloud might pose risks.
Hardware and Software Security for NVIDIA H100 GPUs
- CPU CC Technology: AMD SEV-SNP, Intel TDX CPUs.
- Supported CPUs: AMD Milan (EPYC 7XX3) or AMD Genoa (EPYC 9XX4), Intel Emerald Rapids.
- Supported GPUs: All protections and firewalls enabled on NVIDIA Hopper architecture, including H100 Tensor Core GPUs.
- Software: NVIDIA driver CUDA 12.4, supported hypervisors, and operating systems for AMD and Intel.
Summary
NVIDIA's Confidential Computing solution on H100 Tensor Core GPUs offers a balance between performance and security, ensuring data protection for AI models and sensitive information. With the latest release, including performance optimizations and security enhancements, users can seamlessly transition to confidential, accelerated workloads. For more information, refer to the NVIDIA Trusted Computing Solutions documentation.